TikTok owner ByteDance is actively suppressing a vast number of so-called ‘sensitive words,’ an internal company record reveals.
Internal company records show that TikTok’s parent ByteDance maintains a library of vocabulary lists dictating what users of its platforms can and cannot see. While this is common practice for many social media brands, the Chinese company’s glossary extends beyond the standard moderation blocklists that focus on child safety and things like hate speech.
Posts about the Chinese government, U.S.-China trade, former president Donald Trump, persecuted minority group Uyghurs, and even TikTok competitors like YouTube are subject to monitoring and, in many cases, suppression by ByteDance.
As revealed by Forbes, lists in ByteDance’s tool that tracks “sensitive words” across its platforms feature titles like:
- “TikTok Sensitive Person-Themed Vocabulary”
- “Trump Directed Prohibited Words”
- “Putin Directed Prohibited Words”
- “TikTok Japanese Comments Suppress Words”
- “TikTok Audio Sensitive Words in Tibet Region”
- “Special Prohibited Words for Xi and Peng”
- “Sensitive Words in Douyin Videos in Xinjiang”
- “Theming Strategies of Uyghur-Han Couples”
- “Thematic Strategy for China’s Strategic Policy”
- “Local Life — Taiwan Independence & Hong Kong”
- “YouTube Domestic Surveillance”
- “Company Product Negative Sensitive Vocabulary”
- “TikTok Government Affairs Media Topic Vocabulary”
Forbes reports more than 50 lists in the ByteDance tool with the word “TikTok” or “U.S.” in the title. Internal materials show that TikTok employees accessed the monitoring tool within the last year. Other lists reference TikTok’s Chinese counterpart Douyin and ByteDance products Lark, Toutiao, and Resso.
An internal guide to the tool appears to be written by China-based employees of ByteDance and its Beijing subsidiary Jiyun Hudong. That guide describes its “global core vocabulary” of “common bottom-line sensitive words with high risk” and a “global commonly used thesaurus” containing “fully classified illegal vocabulary.” The document says the system was built in 2017 to help “detect, evaluate, and recall these words across ByteDance products” because they “may bring risks to the company’s safety, reputation, and revenue,” concluding that “the business side shall prevail.”
Jamie Favazza, a spokesperson for TikTok, suggested that the extensive glossary that Forbes obtained could be “significantly outdated or incomplete” and that none of the word lists identified are currently or have ever been used on TikTok. Still, when asked why these lists have “TikTok” or “U.S.” in the titles if they were never used on the platform, Favazza said, “I cannot speculate on the list titles” without Forbes turning over the documents.
Hundreds of documents reviewed by Forbes show a lack of any functional separation internally in access to information, user data, and tools between TikTok and its Chinese counterpart Douyin. Further, dozens of current and former TikTok and ByteDance personnel have revealed that any purported separation between the companies is primarily cosmetic.
In addition to TikTok employees having access to the ByteDance tool in the last year, documents show that ByteDance employees in China are among those who have managed the TikTok tool. Despite this, Favazza says that TikTok could not verify whether any of its employees had accessed the tool without knowing their names.
Records show that these moderation tools and other internal programs collect data on the “hit rate” of sensitive words — including information about U.S. users who post them. One document mentions TikTok and other ByteDance products receiving a recent upgrade that integrated the moderation system with “a new text detection service” that makes it easier to track and analyze “hit records of sensitive words” in real-time. At least one of the engineers running that project was based in Beijing.
Forbes’ findings also revealed extensive overlap between Chinese state media and ByteDance/TikTok employees. Chinese state media has already used TikTok to attempt to influence American users’ opinions about U.S. politics, despite TikTok CEO Shou Zi Chew’s testimony before Congress last month that there was “insufficient” evidence to “prove or disprove that TikTok employs political censorship.”
“The difference is the fundamental issue of ByteDance being subservient to the government and the Ministry of State Security,” says William Evanina, former head of counterintelligence for the U.S. government. “Yes, Facebook and Google all do the same things to protect their global assets on their platforms, but they’re not beholden to the CIA or the NSA. And the NSA and the CIA are not giving Facebook this (glossary).”
From a cybersecurity perspective, experts cite the danger of ByteDance’s monitoring system: the potential to track how often specific words come up, who said or interacted with them, where those people are located, and who else follows them.
In China, laws mandate that companies turn over information to the government if asked. Therefore, ByteDance could be required to share sensitive data with party leaders — whether that activity came from within China, the U.S., or another country.